GitHub Copilot

• Cloud-based AI tool by Github and OpenAI
• Model based on GPT-4 and OpenAI’s Codex
  • Specifically trained on source code
• Basic idea: Plugin for your IDE to integrate Copilot
• Works best for well-represented languages (Python, JS, …)

How to get GitHub Copilot

It’s really easy, but you need:

• GitHub Account
• Active GH Copilot subscription (10$ per month)
  • Get it for free as an academic with an educational account
• IDE that supports Copilot
  • Full support: Visual Studio (Code), Vim, Neovim, JetBrains IDEs (e.g. PyCharm)
  • Limited support: RStudio, ?

See website for step-by-step guide and more information.

Inline code suggestions

• Copilot tries to predict what you want to do next
• Suggestions are based on the context
  • Previous code
  • Comments
  • Variable and function names
  • …

Get better suggestions

• Provide context
  • Open other files
  • Add top level comments explaining the purpose of the script
  • Name variables and functions properly
  • Copy-paste sample code and delete it later
• Be consistent
  • “Garbage in, garbage out”
  • Have a nice and consistent coding style

Nice side effect of using Copilot: More good-practice coding

Chat

• Ask and give commands regarding:
  • Highlighted lines of code
  • The whole script or project
• Preset commands starting with /
  • /fix: fix problems in your code
  • /explain: explain this code
  • /test: write unit tests
  • /new: create new projects or scripts with code

Codium AI as an alternative

• No inline code suggestions
• Great functionality to
  • Explain code
  • Suggestsions improve and enhance code
  • Generate tests
• Not in RStudio, but in VS code and many other IDEs
• Free for personal use (for now)

Concerns to consider

• Privacy
  • Chose whether your prompts and suggestions will be used by Github Github -> Seetings -> Copilot -> Policies
• Plagiarism
  • Block suggestions matching public code (Github -> Seetings -> Copilot -> Policies)
• Ethical concerns
  • For-profit tool trained on open-source
• Environmental concerns
  • Water and enery usage

Usage guidelines

• No definite guidelines, but see examples on lecture website
• Responsibility
  • You are responsible for your scientific output
  • Stay critical, double-check
• Transparency
  • Make clear for which tasks you used which AI
• Know relevant guidelines
  • Journals
  • Your university
• Don’t use AI in exams

Next lecture

Topic t.b.a.

📅 15th February 🕓 4-5 p.m. 📍 Webex

🔔 Subscribe to the mailing list

📧 For topic suggestions and/or feedback send me an email

Thank you for your attention :)

Questions?

References

• Experiment on programmer efficiency with AI tools
• GitHub Copilot
• GitHub Copilot privacy FAQ
• GitHub Copilot Docs: Useful information and guides on how to use Copilot
• Prompt engineering with GitHub Copilot
• Codium AI

References

Guidelines

• DFG Rules on the use of AI particularly for proposals
• Nature living guidelines on responsible use of generative AI in research
• EU AI Act
• Universities (German)
  • FU Berlin “Eckpunktepapier” (German)
  • TU Berlin on AI: Mainly about AI in teaching but contains some general links to other guidelines